A metabolic view on menopause and ageing

Abstract

The ageing of the global population calls for a better understanding of age-related metabolic consequences. Here we report the effects of age, sex and menopause on serum metabolites in 26,065 individuals of Northern European ancestry. Age-specific metabolic fingerprints differ significantly by gender and, in females, a substantial atherogenic shift overlapping the time of menopausal transition is observed. In meta-analysis of 10,083 women, menopause status associates with amino acids glutamine, tyrosine and isoleucine, along with serum cholesterol measures and atherogenic lipoproteins. Among 3,204 women aged 40-55 years, menopause status associates additionally with glycine and total, monounsaturated, and omega-7 and -9 fatty acids. Our findings suggest that, in addition to lipid alterations, menopause may contribute to future metabolic and cardiovascular risk via influencing amino-acid concentrations, adding to the growing evidence of the importance of amino acids in metabolic disease progression. These observations shed light on the metabolic consequences of ageing, gender and menopause at the population level.

ARTICLE

Received 22 Nov 2013 | Accepted 16 Jul 2014 | Published 21 Aug 2014

Kirsi Auro1,2,3,*, Anni Joensuu1,2,*, Krista Fischer4, Johannes Kettunen1,2,5, Perttu Salo1,2, Hannele Mattsson1,2, Marjo Niironen1, Jaakko Kaprio2,6,7, Johan G. Eriksson8,9,10,11,12, Terho Lehtimaki13, Olli Raitakari14, Antti Jula8,

Aila Tiitinen3, Matti Jauhiainen1,2, Pasi Soininen5,15, Antti J. Kangas5,15, Mika Kahnen16, Aki S. Havulinna8, Mika Ala-Korpela5,15,17,18, Veikko Salomaa8, Andres Metspalu4 & Markus Perola1,2,4

The ageing of the global population calls for a better understanding of age-related metabolic consequences. Here we report the effects of age, sex and menopause on serum metabolites in 26,065 individuals of Northern European ancestry. Age-specic metabolic ngerprints differ signicantly by gender and, in females, a substantial atherogenic shift overlapping the time of menopausal transition is observed. In meta-analysis of 10,083 women, menopause status associates with amino acids glutamine, tyrosine and isoleucine, along with serum cholesterol measures and atherogenic lipoproteins. Among 3,204 women aged 4055 years, menopause status associates additionally with glycine and total, monounsaturated, and omega-7 and -9 fatty acids. Our ndings suggest that, in addition to lipid alterations, menopause may contribute to future metabolic and cardiovascular risk via inuencing amino-acid concentrations, adding to the growing evidence of the importance of amino acids in metabolic disease progression. These observations shed light on the metabolic consequences of ageing, gender and menopause at the population level.

1 Public Health Genomics Unit, Department of Chronic Disease Prevention, National Institute for Health and Welfare, Biomedicum 1, Haartmaninkatu 8, Helsinki 00290, Finland. 2 Institute for Molecular Medicine (FIMM), University of Helsinki, Biomedicum 2, Tukholmankatu 8, Helsinki 00290, Finland.

3 Department of Obstetrics and Gynecology, Helsinki University Central Hospital and University of Helsinki, Haartmaninkatu 2, Helsinki 00290, Finland.

4 Estonian Genome Center, University of Tartu, Riia 23b, Tartu 51010, Estonia. 5 Computational Medicine, Institute of Health Sciences, University of Oulu, Pentti Kaiteran katu 1, Oulu 90570, Finland. 6 Departmentof Public Health, Hjelt Institute, University of Helsinki, PO Box 41 Mannerheimintie 172, Helsinki 00014, Finland. 7 Department of Mental Health and Substance Abuse Services, National Institute for Health and Welfare, PO Box 30 (Mannerheimintie 166), Helsinki 00300, Finland. 8 Chronic Disease Epidemiology and Prevention Unit, Department of Chronic Disease Prevention, National Institute for Health and Welfare, Mannerheimintie 166, Helsinki 00300, Finland. 9 Department of General Practice and Primary Health Care, University of Helsinki, PL 20, Tukholmankatu 8B, Helsinki 00029, Finland. 10 Vasa Central Hospital, Sandviksgatan 2-4, Vasa 65130, Finland. 11 Folkhalsan Research Centre, Helsingfors Universitet, PB 63, Helsinki 00014, Finland. 12 Unit of General Practice, Helsinki University Central Hospital, Haartmaninkatu 4, Helsinki 00290, Finland.

13 Department of Clinical Chemistry, Fimlab Laboratories, University of Tampere School of Medicine, Tampere University, Kalevantie 4, Tampere 33014, Finland. 14 Department of Clinical Physiology and Nuclear Medicine, Research Centre of Applied and Preventive Cardiovascular Medicine, Turku University Hospital, University of Turku, Kiinamyllynkatu 4-8, Turku 20521, Finland. 15 NMR Metabolomics Laboratory, School of Pharmacy, University of Eastern Finland, Yliopistonranta 1 PL 1627, Kuopio 70211, Finland. 16 Department of Clinical Physiology, Tampere University Hospital and University of Tampere School of Medicine, Tampere University, Kalevantie 4, Tampere 33014, Finland. 17 Oulu University Hospital, Kajaanintie 50, Oulu 90220, Finland. 18 Computational Medicine, School of Social and Community Medicine and Medical Research Council Integrative Epidemiology Unit, University of Bristol, Senate House, Tyndall Avenue, Bristol, City of Bristol BS8 1TH, UK. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to K.A. (email: mailto:[email protected]

Web End =kirsi.auro@thl. ).

NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

& 2014 Macmillan Publishers Limited. All rights reserved.

DOI: 10.1038/ncomms5708

A metabolic view on menopause and ageing

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708

With global population aging we are facing substantial social, economic and health-care issues. Aging relates closely to a wide spectrum of metabolic

and other chronic diseases. The disease burden signicantly differs by age and sex. Menopause is an endocrinological transition strongly inuencing health and disease susceptibility of all middle-aged and elderly women. It is dened as the permanent cessation of menstruation resulting from the loss of ovarian follicles. Menopause is preceded by menopausal transition, a period of menstrual cycle irregularity, which usually begins in the mid 40s. Menopause is not a distinct event at a discrete point in time but rather a continuous dynamic process of progressive decline in ovarian function and circulating oestrogen levels, with substantial interpersonal age-of-onset variation.

Menopause has substantial health consequences ranging from disturbances in lipid and glucose metabolism to psychological stress and sleep alterations1,2. Importantly, menopause is closely linked with cardiovascular diseases (CVDs), the leading cause of death worldwide, although the signicance of menopause as a CVD risk factor has been challenged36. Premenopausal women have lower CVD incidence and mortality rates compared with men of the same age, which is partly explained by the protective effects of endogenous oestrogens against atherosclerosis7,8. Loss of ovarian function during natural or articial menopause and subsequent hypoestrogenism is believed to promote CVD. Large cohort studies have rmly established increased CVD risk after bilateral oophorectomy, causing sudden menopause9,10. Early menopause may have a modest effect on CVD risk11; however overall, the relationship between postmenopausal status and CVD remains somewhat unclear.

Menopause nevertheless inuences several CVD risk factors, and some menopausal symptoms, such as the prevalence of vasomotor episodes, may reect CVD susceptibility12. Earlier studies reported lower high-density lipoprotein cholesterol (HDL) concentration6 combined with higher low-density lipoprotein cholesterol (LDL)6, triglycerides (TGs) and total cholesterol (TC) concentrations13 among postmenopausal women compared with age-matched premenopausal controls. Later reports suggested LDL cholesterol and apolipoprotein-B (apoB) as age-independent menopausal biomarkers, whereas results on TC, HDL and TG have been controversial14,15. HDL tends to lose its protective effects against CVD after menopause, with an atherogenic shift to higher HDL particle number and smaller HDL particle size16. A recent study suggests an alternative hypothesis, namely that an increased premenopausal cardiovascular risk prole, with higher TC, blood pressure and overweight, promotes by itself early menopause17.

Studies utilizing modern metabolomic approaches have recently proposed novel biomarkers for future insulin resistance18, type II diabetes (T2D)19, CVD20 and mortality21. Metabolome-wide gender22 and age23 issues have been previously assessed as separate entities. We aimed to simultaneously investigate the inuence of age and gender on serum metabolites, further including a key player behind gender differences: menopause. We present data on 26,065 individuals of Northern European ancestry from Finland and Estonia. Assessing a wide spectrum of 135 metabolite measures with highly detailed lipid components sets the scope of metabolomic proling far beyond the traditional laboratory measures and helps to detect minor changes in metabolically important parameters at the individual level. Our results highlight the differences in metabolite ngerprints by gender and show that menopause raises the levels of several lipoprotein and lipid measures and, interestingly, also the levels of amino-acid measures.

ResultsStudy setting. We analysed the effects of age and menopause on systemic metabolomics in altogether 16,107 Finnish and 9,958 Estonian individuals (8,321 and 6,416 women, respectively). The study comprises seven independent Finnish cohorts and a large population cohort from Estonia (EGCUT, Table 1). Individuals using lipid-lowering medication, having diabetes or being pregnant were excluded from all analyses, and individuals using hormone replacement therapy (HRT) were additionally excluded from menopause analyses (Table 1). The metabolites were produced from serum (Finnish) or plasma (Estonian) samples using nuclear magnetic resonance (NMR)-based methods24,25, including well-dened lipid components. The 135 metabolites analysed include 74 lipoprotein, 17 lipid and 23 fatty acid-related measures, accompanied by 9 amino acids and 12 small molecules related to cell energy metabolism (Supplementary Table 1). The Estonian samples lacked information on some lipid and fatty acid components (Supplementary Table 1). Menopause status was dened based on questionnaire data. In addition, follicle-stimulating hormone measures were available in the Finnish cohort DILGOM; the questionnaire data highly correlated with hormone measures in dissecting the menopausal status. All the participants gave written informed consents, and the study cohorts have local ethical board approvals.

A correlation matrix revealed signicant metabolitemetabolite correlations among the 135 metabolites (Supplementary Fig. 1). Principal component (PC) analysis showed altogether 22 PCs explaining 495% of the variation seen in the data set (Supplementary Data 1). The multiple testing corrected signicance level in the highly correlated data was set accordingly (Po0.05/22 PCs, that is, Po0.0023). Further, based on high correlation, only the total particle concentration of each lipoprotein subclass was used in the analyses.

Detailed lipid proles differ by age and gender. To explore age metabolite relations, we constructed a heat map (Fig. 1) of the largest Finnish cohort, FINRISK-97 (n 8,444, age range 2475

years). For each metabolite, the divergence (in units of s.d.) from the mean level of those younger than 35 years was plotted for each 1-year age group. The reference group was chosen to represent a generally healthy population rarely having age-related metabolic disorders. As previous studies have suggested gender differences in the metabolite proles22, a gender-specic approach was chosen. The heat maps revealed clear agesex-related metabolomic ngerprints: among men (Fig. 1; Supplementary Fig. 2a,b), lipoprotein measures of very low-density lipoprotein cholesterol (VLDL), intermediate-density lipoprotein cholesterol (IDL) and LDL, as well as serum total, free and esteried cholesterol, serum TG and apoB all increased starting from the early 30s. The amount of larger HDL particles increased by age, whereas the measures associated with medium and small HDL declined, except for TG in small HDL, which increased. Several metabolites expressed non-linear behaviour by age; this was most evident for the VLDL measures having a parabolic age relation rst increasing but later declining among the oldest men.

The female metabolomic ngerprint (Fig. 1; Supplementary Fig. 2c,d) signicantly differed from the male one. The concentrations of the atherogenic metabolites were relatively stable until a rapid increase occurred at the age of late 40s and early 50s, stabilizing thereafter. Similar to men, VLDL, IDL and LDL concentrations, serum cholesterol measures, larger HDL measures and TG in small HDL all increased by age, and medium and small HDLs declined in the FINRISK-97 women.

2 NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2014 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708 ARTICLE

Table 1 | Characteristics of the study cohorts.

COHORT DILGOM FINRISK-97 HBCS HEALTH 2000 YF FT16 FT12 EGCUT Total Alln 510 8,444 853 1,364 3,596 564 776 9,958 26,065

Womenn 272 4,191 508 781 1,832 278 459 6,416 14,737 Premenopausal/postmenopausal

112/118 2,341/1,167 0/508 151/390 1,832/0 278/0 459/0 3,682/2,641 8,855/ 4824

Diabetes/lipid lower/ HRT

8/22/42 91/87/657 89/82/246 17/69/240 8/0/0 0/0/0 0/0/0 460/228/17 673/448/ 1,202

BMI kg m 2, mean (s.d.)

26.66 (5.36) 26.31 (5.02) 27.22 (4.48) 27.04 (4.79) 24.52 (4.61) 22.18 (3.37) 22.99 (4.03) 26.35 (5.83)

Age in years, mean (range)

51.35 (2574) 47.02 (2474) 61.58 (5669) 57.17 (4574) 31.53 (2439) 26.17 (2330) 22.33 (2024) 45.52 (18103)

Fasting time in hours* (range)

5.26 (118) 5.72 (221) Z8 Z8 Z8 Z8 Z8

Menn 238 4,253 345 583 1,764 286 317 3,542 11,328 Young/oldw 96/142 2,204/2,049 0/345 138/445 1,764/0 286/0 317/0 2,165/1,377 6,970/

4,358

Diabetes/lipid lower 9/48 166/205 57/63 30/69 8/0 0/0 0/0 285/194 555/579 BMI kg m 2, mean(s.d.)

26.66 (5.36) 26.31 (5.02) 27.22 (4.48) 27.04 (4.79) 24.52 (4.61) 22.18 (3.37) 22.99 (4.03) 26.35 (5.83)

Age in years mean (range)

53.45 (2574) 49.74 (2474) 61.30 (5768) 56.92 (4574) 31.36 (2439) 26.21 (2330) 22.43 (2124) 44.76 (1894)

Fasting time in hours* (range)

5.80 (121) 6.32 (323) Z8 Z8 Z8 Z8 Z8 Z4

DILGOM, A subset of FINRISK-07 cohort from Helsinki area; EGCUT, Estonian population sample; FT16, FinnTwin16 cohort; FT12, FinnTwin12 cohort; HBCS, Helsinki Birth Cohort Study; HTR, hormone replacement therapy; YF, Young Finns study.*Minimum or mean fasting time presented.

wYoung o51 years of age, Old Z51 years of age.

The agesex-related patterns in lipids and lipoproteins seen in FINRISK-97 were tested for replication in the Estonian EGCUT cohort (n 9,958, age range 18103 years). Inspection of the

male EGCUT heat map (Supplementary Fig. 3a) showed a similar shift towards atherogenicity in the lipid metabolites as detected among the Finnish men; however, the timing for this change was earlier starting already at the age of 27 years. Especially the VLDL measures expressed a steeper U-shape. Among the Estonian women, again a clear change towards a more atherogenic lipid prole occurred between 45 and 50 years of age (Supplementary Fig. 3b).

Agesex interaction analysis in 40- to 55-year-old participants. The heat maps indicated a major change in the female metabolic ngerprint timing between 45 and 50 years of age, which clearly overlaps the period of menopausal transition. To evaluate the heat map observations, we assessed the genderage effects on metabolites during the menopausal transition period (4055 years, n 3,478) in random effects meta-analysis of three Finnish

cohorts (FINRISK-97, DILGOM and Health2000, containing participants in this age group). We tted a linear model including an age sex interaction term, further adjusted for body mass

index (BMI), daily smoking status and fasting time before sampling. Signicant agesex interactions were observed with 10 metabolites (Table 2). Oesteried cholesterol showed the strongest interaction, followed by lipoprotein particle concentrations of very small VLDL, IDL and medium LDL, total phosphoglycerides, phosphatidylcholine and apolipoprotein A-1 (apoA-1), amino acids glutamine and histidine, and polyunsaturated fatty acids (PUFAs). Figure 2 demonstrates the agesex relations by natural cubic splines, calculated using all the Finnish data (7,786 men and 8,321 women), of the 10 metabolites with signicant agesex interactions, showing the general trend of atherogenic lipoprotein and lipid concentrations increasing and reaching the steady state around the age of 30 years in males and a sudden change around the age of 45 in females. The rest of the VLDL and LDL subclass particle concentration and several fatty

acid concentrations (omega-6, omega-7 and -9 and saturated fatty acids, linoleid acid, docosahexanoid acid, monounsaturated fatty acid (MUFA) and total fatty acids) followed a similar pattern but did not reach the multiple testing corrected signicance level of Po0.0023 (Supplementary Fig. 4; Supplementary Table 2). On the other hand, only minor differences in sex-specic proles were detected in the HDL measures (Supplementary Fig. 4; Supplementary Table 2).

Assessing menopausal effects. We further assessed the impact of menopause status on the metabolite levels in a random effects meta-analysis in female participants of all age groups. Altogether 10,083 Finnish and Estonian women were included in the analysis. The results are presented in Table 3 and Fig. 3. Post-menopausal women had signicantly higher concentrations of serum total, esteried and free cholesterol, apoB, very large and large VLDL particles, all the LDL particle subclasses and IDL particles, and TC in LDL compared with premenopausal women (Table 3A; Fig. 3). Postmenopausal women also had signicantly higher concentrations of amino acids glutamine, tyrosine and isoleucine, as well as sphingomyelins and the ratio of TG to phosphoglycerides (Table 3A; Fig. 3). Potential cohort-induced variability was estimated with the I2 measure of heterogeneity. The mean I2 for the HDL measures was 86.1, for VLDL measures63.0, for LDL measures 68.2 and for fatty acids and related measures 87.1. The amino acids glutamine, tyrosine and isoleucine all had relatively low between-cohort variability (I2o35), whereas greater heterogeneity was seen with glycine, valine and isoleucine (I2480).

We then targeted the model to those women in the menopausal transition window (aged 4055 years, n 3,240), to better focus

on the effects of menopause as opposed to ageing. Postmenopausal women had signicantly higher concentrations of amino acids glutamine and glycine, whereas suggestive association was seen with tyrosine and valine (Table 3B; Fig. 3). Postmenopausal women also had signicantly higher concentrations of MUFA, total and omega-7 and -9 fatty acids (Table 3B; Fig. 3).

NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

& 2014 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708

Lipoproteins Lipids and related measures Amino acids Energy metabolism

Women Age (years) Men Age (years)

24 25 26 27 28 29 30

40

50

31 32 33 34 35 36 37 38 39

41 42 43 44 45 46 47 48 49

51 52 53 54 55 56 57 58 59 60

70

61

71 72 73*

62 63 64 65 66 67 68 69

24 25 26 27 28 29 30

40

50

XXL-VLDL

XL-VLDL

M-VLDL

L-VLDL

S-VLDL

XS-VLDL

IDL

L-LDL

M-LDL

S-LDL

XL-HDL

L-HDL

M-HDL

S-HDL

Triglycerides in VLDL

Cholesterol in LDL

VLDL size

LDL size

Cholesterol in HDL

Cholesterol in HDL2

Cholesterol in HDL3

HDL size

Total cholesterol

Total triglycerides

Esterified cholesterol

Free cholesterol

Apolipoprotein A-1

Apolipoprotein B

ApoB to ApoA1

DHA 22:6

LA 18:2

MUFA

PUFA

Omega-3

Omega-6

Omega-3 to omega-6

Saturated FA

Omega-3 %

Omega-6 %

Saturated FA %

Total fatty acids

CH2 in FA

DB in FA

FA length

BIS to FA

BIS to DB

CH2 per DB

MobCH

MobCH2

MobCH3

Phosphatidylcholine

Sphingomyelins

TotalPG

TG to PG

Alanine

Glutamine

Glycine

Histidine

Isoleucine

Leucine

Phenylalanine

Tyrosine

Valine

Glucose

Glycerol

1-acid glycoprotein

Lactate

Pyruvate

Citrate

Acetate

Acetoacetate

3-hydroxybutyrate

Albumine

Creatinine

Urea

31 32 33 34 35 36 37 38 39

41 42 43 44 45 46 47 48 49

51 52 53 54 55 56 57 58 59 60

70

61

71 72 73*

62 63 64 65 66 67 68 69

Metabolites

2 1 0 1 2

S.d.

Figure 1 | Heat maps demonstrating age-related changes in the metabolites in 3,009 FINRISK-97 women (upper panel) and 3,450 men (lower panel). One-year age groups are displayed on the y axis and individual metabolites on the x axis. Colours represent the divergence in units of s.d. from the mean of the metabolite levels of 2435 year olds. Heat maps of the Estonian cohort are presented in the Supplementary Fig. 3. The horizontal lines are set at the mean age of menopause, 51 years. Lipoprotein subclasses represent particle concentration of each subclass. VLDL very low-density lipoprotein,

IDL intermediate-density lipoprotein, LDL low-density lipoprotein, HDL high-density lipoprotein, ApoB apolipoprotein-B, ApoA-1 apolipoprotein

A-1, DHA docohexanoid acid, LA linoleic acid, MUFA monounsaturated fatty acids, PUFA other polyunsaturated fatty acids than 18:2, FA fatty

acids, CH2 methylene groups, DB double bonds, BIS bisallylic groups, MobCH double bond protons of mobile lipids, MobCH2 CH2 groups of

mobile lipids, MobCH3 CH3 groups of mobile lipids, PG phosphoglycerides, TG triglycerides.

4 NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2014 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708 ARTICLE

Table 2 | Agesex interaction on metabolite levels in random effects meta-analysis among 40- to 55-year-old Finnish men and women.

Metabolite b (CI 95%) P value I2 Esteried cholesterol 0.042 (0.0270.057) 2.29E 08 35.3

Concentration of very small VLDL particles 0.036 (0.0220.050) 8.08E 07 0

Concentration of IDL particles 0.035 (0.0210.049) 8.34E 07 4.5

Total phosphoglycerides 0.034 (0.0190.050) 1.06E 06 0.1

Glutamine 0.034 (0.0190.049) 6.35E 06 34.0

Histidine 0.035 (0.0190.050) 8.41E 06 0.1

Phosphatidylcholine 0.032 (0.0170.047) 2.60E 05 0.2

Apolipoprotein A-1 0.028 (0.0130.043) 0.000239 0 Concentration of medium LDL particles 0.031 (0.0140.047) 0.000267 34.8 Polyunsaturated fatty acids 0.038 (0.0170.059) 0.000485 0

BMI, body mass index; CI, condence interval; IDL, intermediate-density lipoprotein; LDL, low-density lipoprotein; VLDL, very low-density lipoprotein.

Figures are from random effects meta-analysis using multivariate linear regression testing age sex interaction effect on metabolite concentrations. Three Finnish cohorts were included (FINRISK-97,

DILGOM and Health2000). The linear model is adjusted for BMI, daily smoking and fasting time before sampling.

Z-score of esterified

cholesterol concentration

Z-score of total

phosphoglyceride concentration

Esterified cholesterol

Concentration of very

small VLDL particles

Concentration of IDL particles Glutamine Histidine

2 1 0 1 2

20 30 40 50 60 70 20 30 40 50 60 70 20 30 40 50 60 70 20 30 40 50 60 70 20 30 40 50 60 70

Age in years Age in years Age in years Age in years Age in years

20 30 40 50 60 70 20 30 40 50 60 70 20 30 40 50 60 70 20 30 40 50 60 70

Age in years Age in years Age in years Age in years

Z-score of very small VLDL

particle concentration

Z-score of IDL

particle concentration

Z-score of medium LDL

particle concentration

2 1 0 1 2

Z-score of glutamine

concentration

2 1 0 1 2

Z-score of histidine

concentration

2 1 0 1 2

2 1 0 1 2

Total phosphoglycerides Phosphatidylcholine Concentration of

mediun LDL particles

Apolipoprotein A-1 Polyunsaturated fatty acids

2 1 0 1 2

Z-score of phosphatidyl

choline concentration

2 1 0 1 2

2 1 0 1 2

Z-score of apolipoprotein

A-1 concentration

2 1 0 1 2

Z-score of polyunsaturated

fatty acids concentration

20 30 40 50 60 70

Age in years

2 1 0 1 2

Figure 2 | Illustration of the key menopause-associating metabolite concentrations by age t with natural cubic splines. Altogether, 6,790 Finnish men (blue line) and 7,544 women (red line) were analysed with natural splines, utilizing univariate regression, using z-standardized values (individual values subtracted from the mean and divided by s.d.). Knots are placed at every 5 years. Standardization was performed separately in men and women; thus, the absolute values are not comparable between sexes. P values outline age sex interaction effects on the metabolite concentrations,

calculated using multivariate liner regression in random effects meta-analysis (see also Table 2).

In addition, signicant menopausal associations were seen with free and esteried cholesterol and the apoB/apoA-1 ratio (Table 3B; Fig. 3). Postmenopausal women had less bisallylic groups per double bond than premenopausal women. The lipoprotein subclass concentrations showed a similar trend as among all women; however, the results did not reach statistical signicance (Fig. 3; Supplementary Table 3). The HDL measures (mean I2 2.7), fatty acids and related measures (mean

I2 10.5), and amino acids glutamine, tyrosine, isoleucine and

valine (I2o27) had relatively low cohort variability, whereas mean I2 for LDL measures was 37.1 and for VLDL measures 44.0.

Sensitivity analyses. As all women using HRT were excluded from the analyses because of uncertain menopausal status, we performed a sensitivity analysis on possible HRT effects on metabolomic ngerprint. Finnish 40- to 60-year-old HRT users (n 1,053) were compared with premenopausal non-users

(n 813) and separately with postmenopausal non-users

(n 735) of the same age group in random effects meta-analysis

(Supplementary Table 4). The metabolomic prole of HRT users did not differ signicantly from the premenopausal non-users

with any of the metabolites, whereas the postmenopausal non-users had pro-atherogenic changes in their lipid prole compared with the HRT users in terms of higher values of ApoB and ApoB/ApoA-1 ratio, and higher concentration of very small and very large VLDL particles, large LDL particles and IDL particles, and lower concentrations of large HDL particles, lower TC in HDL and HDL2, as well as smaller diameter of HDL particles. Postmenopausal non-users also had higher values of albumin and glycoprotein acetyls than the HRT users.

Sensitivity analyses were also performed to assess potential impacts of fasting time variability as well as age effects. Random effects meta-analysis among Finnish 40- to 55-year-old women (n 1,804) revealed signicant age effects with several

metabolites, including amino acids, lipids and lipoproteins (Supplementary Table 5). Fasting time was assessed accordingly (n 1,490); however, signicant inuence was seen only with

four metabolites (Supplementary Table 6).

The HDL measures, measures describing fatty acid saturation and measures of small molecules related to cell energy metabolism did not show consistent association with menopause status or an agesex interaction (Supplementary Tables 2 and 3).

NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

& 2014 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708

Table 3 | The association of the metabolites with menopause status in random effects meta-analysis using ultivariate linear regression in (a) 21- to 75-year old (n 10,083) and (b) 40- to 55-year-old (n 3,240) Finnish and Estonian women.

A B

Metabolite b (CI 95%) P value I2 Metabolite b (CI 95%) P value I2 Free C 0.39 (0.290.48) 1.1e 14 0 MUFA 0.29 (0.130.44) 0.00013 0

Sphingomyelins 0.34 (0.240.45) 1.4e 10 4.3 Free C 0.29 (0.130.44) 0.00026 4.6

Glutamine 0.24 (0.160.31) 2.0e 10 5.2 Glycine 0.26 (0.120.41) 0.00042 0

Tyrosine 0.20 (0.130.27) 2.8e 08 0 Glutamine 0.19 (0.0830.30) 0.00053 21.3

TG/PG 0.28 (0.160.40) 8.0e 08 16.2 o-7 and -9 fatty acids 0.23 (0.0920.37) 0.0011 0

ApoB 0.25 (0.120.38) 0.00014 59.9 ApoB/ApoA-1 0.15 (0.0600.24) 0.0011 0.3 Large LDL 0.24 (0.110.36) 0.00019 58.1 Bis DB 0.24 ( 0.38 to 0.091) 0.0014 0

Very large VLDL 0.28 (0.130.43) 0.00019 71.3 Esteried C 0.23 (0.0880.38) 0.0016 0 Esteried C 0.41 (0.190.64) 0.00027 69.6 Total fatty acids 0.22 (0.0840.36) 0.0017 0 Total LDL 0.21 (0.100.33) 0.00034 51.1IDL 0.25 (0.110.40) 0.00045 68.2Medium LDL 0.21 (0.090.33) 0.00052 54,6Isoleucine 0.23 (0.100.37) 0.00070 59.1Small LDL 0.20 (0.080.32) 0.0011 57.7Serum C 0.36 (0.140.58) 0.0014 86.9Large VLDL 0.24 (0.090.39) 0.0016 69.2

ApoB, apolipoprotein-B; ApoB/ApoA-1, apolipoprotein B/apolipoprotein A-1; Bis DB, Bisallylic groups per double bonds; C, cholesterol; CI, condence interval; IDL, intermediate-density lipoprotein; LDL, low-density lipoprotein; MUFA, monounsaturated fatty acids; o, Omega; TG/PG, triglycerides/phosphoglycerides; VLDL, very low-density lipoprotein.

The measures of IDL, LDL and LVDL refer to the stated particle concentration.

The concentrations of TC in HDL, HDL2 and HDL3 particles also remained relatively stable despite ageing, whereas small and medium HDL concentrations declined by age in both sexes, and very large HDL increased (Supplementary Fig. 4).

DiscussionWe assessed the effects of age, sex and menopause on metabolites among 426,000 Finnish and Estonian participants of Northern

European ancestry. The study provides examples of gender-specic metabolic ngerprints at the population level, showing association of menopause with cholesterol components, amino acid and fatty acid concentrations, and an evident pattern of proatherogenic lipid changes. Among men, the concentrations of these lipid measures rise already at early middle age. Premenopausal women have signicantly lower mean values until a rapid increase around the age of 45, where the atherogenic lipid concentrations suddenly bounce, bypassing the mean values of men. This pattern is consistent for all the cholesterol measures analysed, all the VLDL, IDL and LDL subclass particle concentrations, as well as for TG measures and apoB.

Menopause-associated oestrogen depletion has several well-established metabolic consequences. Menopause promotes atherosclerosis through changes in lipid prole, endothelial dysfunction and oxidative stress26,27. Hormonal changes in menopause promote higher body fat, in particular abdominal obesity, contributing to the increased CVD and metabolic risk28,29. Hormone levels do not, however, independently predict dyslipidemias or CVD events30. Menopause is most likely one of the key players behind the substantial differences observed here between the male and female age-specic metabolic ngerprints. High TC, LDL, TG and apoB values have also previously been associated with menopause14,15,29,31. These lipid disturbances may be especially harmful in hyperglycaemic environment promoting oxidative injury31,32.

Recent studies have targeted to more detailed lipid proles, beyond the traditional measures of TC, total LDL cholesterol or TG. Menopause may affect the concentrations of small, dense LDL alone and with apoB33. VLDL size and IDL particle number have been proposed as predictors for stroke in postmenopausal women34, and both small, dense LDL and small-sized HDL have

displayed predictive value for future T2D in women32. Our results provide insights into highly detailed lipid subclass measures by age and sex in a population free of diabetes and not using lipid-lowering medication. We observed signicant associations of free and esteried cholesterol and the apoB/apoA-1 ratio with menopause status among 40- to 55-year-old women, postmenopausal women having signicantly higher values. Esteried cholesterol, very small VLDL, IDL, medium LDL and apoA-1 were among the key ndings in the agesex interaction analysis, suggesting a signicant gender inuence on these measures. These ndings also highlight the role of menopause as a CVD risk factor. Via pro-atherogenic lipid alterations menopause signicantly contributes to CVD risk.

Phosphoglycerides, sphingomyelins and phosphatidylcholine and other cholines, all related to lipoprotein assembly, also showed signicant agesex interaction or association with menopause status. The menopause-induced increase in these lipoproteins, lipids and fatty acids likely reects the inuence of hormonal alterations on liver enzymes35,36 but might also relate to weight gain and insulin resistance37,38. The HDL measures analysed did not show consistent association with menopause or an agesex interaction, being in line with the emerging understanding of absolute HDL concentration being a poor indicator of HDL functional capacity39.

The relation of obesity, insulin secretion and branched-chain amino acids has been studied for decades40. Recently, branched-chain and aromatic amino acids, leucine, isoleucine, valine, tyrosine and phenylalanine have been linked to insulin resistance17 and risk of future T2D18, some in an obesity-dependent manner17. A risk score of amino acids tyrosine, phenylalanine and isoleucine has been shown to predict future T2D19 and CVD20 in healthy individuals. Glutamine and glutamic acid metabolism may contribute to the regulation of glucose metabolism and insulin secretion in diabetic patients in a protective manner4143. High glutamine-to-glutamate ratio reduced the risk of incident diabetes in mice44. On the other hand, parenterally or enterally administrated glutamine is suggested to have cardioprotective effects in patients undergoing cardiac surgery45,46. Qi et al.47 have recently established a novel risk locus for coronary heart disease among diabetic patients, located in the GLUL gene. The mechanisms by which alterations

6 NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2014 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708 ARTICLE

Women aged 4055 Women aged 2575

XXL

XL

L

M

S

XS

L

M

S

XL

L

M

S

Triglycerides in VLDL

VLDL size

Cholesterol in LDL Cholesterol in HDL

Cholesterol in HDL2

Cholesterol in HDL3 Total cholesterol

Total triglycerides

Esterified cholesterol

Free cholesterol

Apolipoprotein A-1

Apolipoprotein B

ApoB to ApoA1

DHA 22:6

LA 18:2

MUFA

PUFA

Omega-3

Omega-6

Omega-3 %

Omega-6 %

P < 0.0023

P < 0.05

VLDL

Lipoproteins Lipids and related measures Amino acids Energy metabolism

Concentration of

IDL LDL

HDL

LDL size

HDL size

Omega-3 to omega-6

Saturated FA

Saturated FA %

Total fatty acids

CH2 in FA

DB in FA

FA length

BIS to FA

BIS to DB

CH2 per DB

MobCH

MobCH2

MobCH3

Phosphatidylcholine

Sphingomyelins

Total PG

TG to PG

Alanine

Glutamine

Glycine

Histidine

Isoleucine

Leucine

Phenylalanine

Tyrosine

Valine

Glucose

Glycerol

1-acid glycoprotein

Lactate

Pyruvate

Citrate

Acetata

Acetoacetate

3-hydroxybutyrate

Albumine

Creatinine

Urea

1.0 0.6 0.2 0.2 0.6 1.0 1.0 0.6 0.2 0.2 0.6 1.0

Effect of menopause and 95% CIs Effect of menopause and 95% CIs

Figure 3 | The association of the metabolites with menopause status in Finnish and Estonian women in random effects meta-analysis using multivariate linear regression. The effect sizes and 95% condence intervals are presented. Values from linear regression model are adjusted for age, BMI, smoking and time of fasting before sampling. On the left side, results of the women in the menopausal transition window (aged 4055 years, n 3,204),

and results of all women (2175 years, n 10,083) on the right side of the gure. Dark-red colour indicates a P value below the Bonferroni corrected

signicance level (Po0.0023), whereas light-red colour indicates Po0.05. b indicates the effect size in s.d. within the rank-normalized values. VLDL very

low-density lipoprotein, IDL intermediate-density lipoprotein, LDL low-density lipoprotein, HDL high-density lipoprotein, ApoB apolipoprotein-B,

ApoA-1 apolipoprotein A-1, DHA docohexanoid acid, LA linoleic acid, MUFA monounsaturated fatty acids, PUFA other polyunsaturated fatty

acids than 18:2, FA fatty acids, CH2 methylene groups, DB double bonds, BIS bisallylic groups, MobCH double bond protons of mobile lipids,

MobCH2 CH2 groups of mobile lipids, MobCH3 CH3 groups of mobile lipids, PG phosphoglycerides, TG triglycerides.

NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

& 2014 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708

in the metabolism or concentration of glutamine, glutamate or their precursors may increase CVD risk are unclear. At the population level in humans, metabolomics proling with mass-spectrometry data has linked several amino acids, including glutamine and isoleucine, with glucose intolerance and T2D48, and glutamine concentration, has been positively associated with higher intimamedia thickness of carotid artery as well as with coronary artery atherosclerosis49. The same study also linked tyrosine concentrations with these two CVD-related end points. Here postmenopausal women had higher glutamine, tyrosine and isoleucine concentrations than premenopausal women, suggesting a role for menopause in their regulation. Among 4055-year-old women, the values of glutamine and glycine were signicantly higher in postmenopausal than in premenopausal women, and tyrosine and valine showed suggestive associations. Menopause may thus contribute to the risk of cardiometabolic end points by affecting amino-acid levels, in addition to lipid alterations and obesity. The associations of amino-acid concentrations with menopause have not been previously reported in large population samples.

Fatty acid composition and intake modulate CVD risk in several ways. Omega-3 PUFAs act cardioprotectively via anti-inammatory and blood pressure-lowering actions50. Higher PUFA intake is associated with lower CVD incidence and mortality51,52. Docosahexaenoic acid, belonging to omega-3 long-chain PUFA, is proposed to have TG-lowering capacity53. Trans-fatty acid concentrations54 and high intake51 are linked with increased CVD risk, whereas long-chain MUFA are linked with the risk of congestive heart failure55. Menopause affects fatty acid metabolism at least through obesity-related changes and altering tissue fatty acid metabolism. Oestrogen affects tissue PUFA content56 and the capacity of adipose tissue to uptake and store fatty acids in TG-enriched lipid droplets57. In postmenopausal women, PUFA intake improved endothelial function27 and physical performance58. Fatty acids also play a role in ageing; small-to-medium-chain saturated fatty acids were inversely associated with telomere length in postmenopausal women in a recent study59. Our observations with MUFA, total and omega-7 and -9 fatty acids associating with menopause status and signicant agesex interaction seen with PUFA contribute to the current understanding of the menopausal inuence on CVD-related pathways. The same trend in agesex interactions was also seen with several other fatty acids (omega-6, omega-7 and -9 fatty acids, docosahexaenoid acid, and linoleic acid). Unfortunately, nutritional data on fatty acid intake were not available to assess the role of diet in the observed menopausal associations.

A thorough assessment of premenopause, perimenopause and menopause would require highly detailed targeting of the nal menstrual period and the appearance of the preceding menopause-related symptoms, including longitudinal hormone and risk factor measures before, during and after the menopausal transition60. The vast majority of large population studies lack such information, relying on questionnaire data and being cross-sectional in nature. Factors such as surgical menopause and the wide use of HRT and other hormonal agents masking the postmenopausal hallmarks further blur the recognition of natural menopause. In addition, menopause-related hormonal alterations and hence the potential menopausal effects on, for example, CVD risk factors begin already years before the nal menstrual period or any notable symptoms61. We veried the questionnaire-based menopausal data were partly using hormone measurements, and the two were very well in line with each other. Reliable recognition of perimenopausal women was however impossible, and all the women using HRT were excluded as menopause status was analysed as a dichotomous end point. A sensitivity analysis

comparing HRT users with pre- and postmenopausal non-users suggested that the metabolomic prole of the HRT users resembled that of premenopausal women, whereas the postmenopausal non-users had pro-atherogenic changes in their lipid prole compared with the HRT users.

The importance of menopause as a CVD risk factor is controversial. While articial menopause by bilateral oophorectomy has been associated with increased CVD risk9,10, large cohort studies have failed to establish the link between menopause and CVD36. Menopause does, however, inuence several CVD risk factors. Our results link menopause with amino-acid alterations, which in turn have been linked with CVD20, insulin resistance17 and T2D1820 in separate, large population studies. Individual lipid markers such as LDL and HDL cholesterol have been shown to predict CVD since the 1950s, and menopause inuences these lipid traits in a proatherogenic manner1416. Here menopause had a pro-atherogenic inuence on several lipids and lipoprotein subclasses. Postmenopausal women not using HRT also had signs of proatherogenic lipid prole compared with the HRT users. These ndings elucidate the potential impact of menopause in cardiometabolic processes.

A major strength of this study is the substantial sample size of over 26,000 participants. The metabolites were measured in the same laboratory. Limitations of this work include the cross-sectional design, allowing crude estimates of menopausal status. Assessment of menopause as a continuous variable ranging from premenopause to perimenopause to late postmenopause, more closely reecting the underlying biology, was however beyond the reach of the questionnaire data. Age of the participants varied from 21 to 103 years. Heat maps, used to assess the age effects, revealed signicant age effects on several metabolites, which were also detected in sensitivity analysis. Nevertheless, menopause status still had a signicant impact on several metabolites. In FINRISK-97, DILGOM and EGCUT cohorts, only semi-fasting state was required before sampling. Although all analyses were corrected for age and fasting time, variances between the cohorts analyzed may have an impact on the results. The potential cohort-induced variation was controlled for using meta-analysis-based methods. The heterogeneity seen in random effects meta-analysis may hint on age or sex inuence, or reect differences in fasting time. Relatively low between-cohort variance was seen with amino acids glutamine, isoleucine and tyrosine, and these amino acids may be less sensitive to age and sex effects than amino acids glycine, valine and phenylalanine. The same cohorts have been utilized in a similar study setting previously21,25,62. The study comprised individuals of Northern European ancestry, and the ndings may vary in other populations. Nutritional data on fatty acid intake were not available to evaluate the potential nutritional inuence on menopause-related ndings.

Metabolites were analysed from serum samples in all the Finnish study cohorts, but from plasma in the Estonian study sample, which may produce some variability across the cohorts, although the correlation between serum- and plasma-based values has been reported to be high for the majority of the metabolites63. Serum metabolome comprises thousands of components related to systemic metabolism, of which NMR-based applications only capture a small fraction. Nevertheless, NMR makes metabolomics available for the masses and has facilitated a variety of novel scientic ndings. Currently, NMR-based applications are the only approaches that can offer fully automated and highly reproducible high-throughput experimentation in a very cost-effective manner64. Cost effectiveness via high-throughput, along with absolute quantication of molecular measures, is a prerequisite for metabolomics methodologies in epidemiology.

8 NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2014 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708 ARTICLE

Further associations of menopause-induced metabolite alterations with diagnostic disease end points, such as CVD or T2D, should be evaluated in future studies, thus widening the biological relevance of the results presented here. Despite these limitations, our ndings unravel age and menopausal effects on serum metabolome, adding to the current understanding of menopause-induced metabolic changes.

To summarize, our results demonstrate that metabolic ngerprints are strongly age- and sex-dependent. Men enter the pro-atherogenic phase already in the early middle age. In women, a clear atherogenic shift overlapping the menopausal transition is observed. Our ndings of a menopause-specic increase regarding several amino acids and fatty acids suggest a menopausal inuence on numerous pathways contributing to the pathogenesis of diabetes and of CVDs.

Methods

Study cohorts. This study comprises seven independent Finnish study cohorts and an Estonian population sample, totalling 26,065 individuals (Table 1). All the study cohorts have been previously described in detail6568. FINRISK cohort samples are collected every 5 years in ve separate geographical areas in Finland. In each survey 8,00010,000 persons aged 2475 years are invited to ll in questionnaires and to participate in a clinical examination and blood tests. The FINRISK-97 sample was collected in 1997 and had 8,444 participants. The DILGOM sample is a subset of the FINRISK-07 collection with 555 individuals from the Helsinki region examined in great detail for metabolic risk factors. Health2000 is an independent study sample with 8,223 participants from Finland. Metabolome data were available for 1,364 participants belonging to a subset with detailed CVD risk factor measures, all Z45 years. The Young Finns (YF) study assesses CVD risk factors among 3,500 young individuals, who at baseline were 318 years of age. Serum samples for the metabolomics analyses were drawn in 2001, when participants were 2439 years of age. Helsinki Birth Cohort Study (HBCS) comprises participants born in 19341944 in Helsinki, Finland. In HBCS, metabolome data was available for 853 individuals. FinnTwin16 (FT16) is a longitudinal twin cohort study with 5,563 Finnish twins born 19741979. FinnTwin12 (FT12) sample consists of 5,362 Finnish twins born in 19831987 followed longitudinally. Metabolome data were available for 554 and 776 participants, respectively, collected during a clinical study in young adulthood. The EGCUT sample is an Estonian population cohort with metabolome data of 9,958 participants aged 18103 years. All cohorts have detailed questionnaire data with medical examination at the time of metabolome serum sampling. All participants gave written informed consent, and all studies have local ethical committee approvals (the ethics committee of Finnish National Health Institute for FINRISK-97 (38/96) and Health2000 (407/E3/2000), the ethics committee of Helsinki University Hospital, Finland (IRB00003181), for DILGOM (229/E0/2006), the twin cohorts (346/E0/05 and 113/E3/2001) and HBCS (334/E3/2000), the 1st ethical committee of Southwestern Hospital District, Finland, for the YF study (3/2000), and the ethics committee of the University of Tartu, Estonia, for EGCUT (205/T-20).

Serum metabolites. The same high-throughput 1H NMR metabolomics platform in the same laboratory was used to produce all the metabolite variables used in the study, including the Estonian ones. The method has been described in detail elsewhere24,25,62. The NMR-based metabolome technique yields data on 135 serum metabolites and ratios (Supplementary Table 1). These include 74 lipoprotein subclass measures, 17 lipid measures, 23 fatty acids and measures related to fatty acid saturation, 9 amino acids and 12 small molecules, many involved in cell energy metabolism. The metabolites were produced from serum in all the Finns and from plasma in Estonians.

Assessing menopause. Menopausal status was assessed rst by dening participants aged 35 or younger as premenopausal and those aged 56 years or more as postmenopausal, unless otherwise stated in the questionnaire, for example, due to surgical menopause. Questionnaire data were then used to distinguish the meno-pause status for the remaining individuals. All women o55 years reporting regular menstruation without hormonal agents were classied as premenopausal, and all women 435 years reporting having their nal menstrual period more than 1 year ago were stated postmenopausal, together with all women who had undergone surgical menopause. HRT users and participants with missing or incomprehensive menstrual data were excluded due to uncertain menopausal status (n 1,202).

Hormone measures or longitudinal data on menstruation were not available. In addition, we excluded pregnant women and those men and women using diabetes medication or statins in order to control for confounding factors (n 1,248 in

total). Follicle-stimulating hormone measures from serum (S-FSH, IU l 1) were available for 271 women in the DILGOM study cohort and were used to verify the questionnaire data. Postmenopausality limit of S-FSH Z26 IU l 1, recommended

by the Helsinki University Central Hospital laboratory, was utilized. The correlation of the questionnaire results and hormone-based menopausal approximation was extremely high: only three women were determined postmenopausal according to the questionnaire but had S-FSH o15 IU l 1 each, and two women classied as premenopausal had S-FSH426 IU l 1.

Statistical analyses. R (version 2.12 or higher, http://www.r-project.org/

Web End =http://www.r-project.org/ ) was used for all statistical analysis.

Correlation matrix and PCs. Pairwise Spearmans rank correlation coefcients were calculated from the metabolite concentrations of the combined data set of the Finnish cohorts (n 12,629) with the cor.test function (method spearman) and

plotted with the heat map.2 function in the gplots package of R. The correlation matrix is presented in Supplementary Fig. 1. PCs were calculated utilizing all the Finnish data with the princomp-function of R. The rst 22 components explained 495% of the variation, of which the six rst components explained 480% of the variation in the data set (Supplementary Data). The remaining 16 components explained only o2% of variation each. On the basis of PCs, the multiple testing corrected signicance level was set to Po0.05/22 components Po0.0023.

The PCs did not follow clear physiological categories.

Heat map of metabolite levels. To visualize the change in metabolite concentrations by age in FINRISK-97, the mean of the standardized metabolite level per 1-year age group (age, age 1) was plotted for each metabolite and sex.

Metabolite standardization and plotting were performed separately for males and females. In the standardization, the mean of under 35-year olds was used as reference instead of the mean of all observations to account for the dissimilarities in age distributions between the sexes (mean age 45.1 in FINRISK-97 females and48.8 in males) and to compare against values of metabolically healthy individuals. The change in colour represents, thus, the divergence in units of s.d.s from the mean of the metabolite levels of 2435-year olds. Clear individual metabolite outlier values (from 13 subjects in FINRISK-97) were removed before metabolite standardization to prevent individual age groups from standing out due to outliers. The heat maps were plotted in a similar manner in the Estonian EGCUT cohort with 2435 year olds as the reference age group.

Agesex interaction and splines. Agesex interactions were assessed with an age sex interaction term in a random effects meta-analysis (rma.uni-function in

metafor package of R, mean I2 35.2) using a linear regression model, adjusted

further for BMI, daily smoking status (as a dichotomous variable), and fasting time (hours) before sampling ([metaboliteBage sex BMI daily smoking fasting

time). Random effects model was employed in all analyses because of differences in the characteristics of cohorts used, as seen in Table 1, and for better comparison between different metabolites. Men and women aged 4055 years, that is, the age overlapping the menopausal transition period, from three Finnish cohorts (FINRISK-97, DILGOM and Health2000, comprising participants in the selected age group, n 3,478) were included in the analysis. Metabolites were normalized

separately by rank-based inverse normal transformation within each cohort. Natural cubic splines were constructed of all the Finnish participants with the ns-function of the splines package of R. Splines are based on the s.d.-normalized score of the metabolite concentrations with knots placed every 5 years of age.

Menopause analyses. Menopausal effects were explored with random effects meta-analysis using the restricted maximum-likelihood estimator approach68,69 (rma.uni-function with method REML in metafor package of R) utilizing70

a linear regression model (metaboliteBmenopause status age BMI daily

smoking fasting time). The FINRISK-97 and the DILGOM cohorts were

additionally adjusted for the geographical eastern-western status. Menopause status was assessed as a dichotomous variable. To achieve normality, rank-based inverse normal transformation was applied to the metabolite concentrations, to attain the means of zero and s.d.s of one. The rst phase comprised 10,083 women from Finland and Estonia (age range 2175 years, mean I2 73.5). The analysis was then

further targeted to those women in the menopausal transition window (aged 4055 years, n 3,204, mean I2 27.2). The I2 gures are from meta-analysis performed

as stated above.

References

1. Rosano, G. M., Vitale, C., Marazzi, G. & Volterrani, M. Menopause and cardiovascular disease: the evidence. Climacteric 10, 1924 (2007).

2. Nelson, H. D. Menopause. Lancet 371, 760770 (2008).3. Tracy, R. E. Sex difference in coronary disease: two opposing views. J. Chronic Dis. 19, 12451251 (1966).

4. Furman, R. H. Are gonadal hormones (estrogens and androgens) of signicance in the development of ischemic heart disease. Ann. N. Y. Acad. Sci. 149, 822833 (1968).

5. Heller, R. F. & Jacobs, H. S. Coronary heart disease in relation to age, sex, and the menopause. Br. Med. J. 1, 472474 (1978).

NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

& 2014 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708

6. Tunstall-Pedoe, H. Myth and paradox of coronary risk and the menopause. Lancet 351, 14251427 (1998).

7. Mendelsohn, M. E. & Karas, R. H. Molecular and cellular basis of cardiovascular gender differences. Science 308, 15831587 (2005).

8. Mikkola, T. S., Gissler, M., Merikukka, M., Tuomikoski, P. & Ylikorkala, O. Sex differences in age-related cardiovascular mortality. PLoS ONE 8, e63347 (2013).

9. Colditz, G. A. et al. Menopause and the risk of coronary heart disease in women. New Engl. J. Med. 316, 11051110 (1987).

10. Matthews, K. A. et al. Menopause and risk factors for coronary heart disease. New Engl. J. Med. 321, 641646 (1989).

11. Atsma, F., Bartelink, M. L., Grobbee, D. E. & van der Schouw, Y. T. Postmenopausal status and early menopause as independent risk factors for cardiovascular disease: a meta-analysis. Menopause 13, 265279 (2006).

12. Tuomikoski, P., Ylikorkala, O. & Mikkola, T. S. Menopausal hot ushes and vascular health. Ann. Med. 43, 283291 (2011).

13. Lindquist, O. Intraindividual changes of blood pressure, serum lipids, and body weight in relation to menstrual status: results from a prospective population study of women in Gteborg, Sweden. Prev. Med. 11, 162172 (1982).14. Matthews, K. A. et al. Are changes in cardiovascular disease risk factors in midlife women due to chronological aging or to the menopausal transition?J. Am. Coll. Cardiol. 54, 23662373 (2009).15. He, L. et al. Menopause with cardiovascular disease and its risk factors among rural Chinese women in Beijing: a population-based study. Maturitas 72, 132138 (2012).

16. Woodard, G. A. et al. Lipids, menopause, and early atherosclerosis in Study of Womens Health Across the Nation Heart women. Menopause 18, 376384 (2011).

17. Kok, H. S. et al. Heart disease risk determines menopausal age rather than the reverse. J. Am. Coll. Cardiol. 47, 19761983 (2006).

18. Wrtz, P. et al. Metabolic signatures of insulin resistance in 7,098 young adults. Diabetes 61, 13721380 (2012).

19. Wang, T. J. et al. Metabolite proles and the risk of developing diabetes. Nat. Med. 17, 448453 (2011).

20. Magnusson, M. et al. A diabetes-predictive amino acid score and future cardiovascular disease. Eur. Heart J. 34, 19821989 (2013).

21. Fischer, K. et al. Biomarker proling by nuclear magnetic resonance spectroscopy for the prediction of all-cause mortality: an observational study of 17,345 persons. PLoS Med. 11, e1001606 (2014).

22. Mittelstrass, K. et al. Discovery of sexual dimorphisms in metabolic and genetic biomarkers. PLoS Genet. 7, e1002215 (2011).

23. Yu, Z. et al. Human serum metabolic proles are age dependent. Aging Cell 11, 960967 (2012).

24. Soininen, P. et al. High-throughput serum NMR metabonomics for cost-effective holistic studies on systemic metabolism. Analyst 134, 17811785 (2009).

25. Kettunen, J. et al. Genome-wide association study identies multipleloci inuencing human serum metabolite levels. Nat. Genet. 44, 269276 (2012).

26. Wassmann, S. et al. Endothelial dysfunction and oxidative stress during estrogen deciency in spontaneously hypertensive rats. Circulation 103, 435441 (2001).

27. Gortan Cappellari, G. et al. Treatment with n-3 polyunsaturated fatty acids reverses endothelial dysfunction and oxidative stress in experimental menopause. J. Nutr. Biochem. 24, 371379 (2013).

28. Davis, S. R. et al. Understanding weight gain at menopause. Climacteric (2012; 15, 419429 (2012).

29. Abdulnour, J. et al. The effect of the menopausal transition on body composition and cardiometabolic risk factors: a Montreal-Ottawa New Emerging Team group study. Menopause 19, 760767 (2012).

30. Chen, Y. et al. Endogenous hormones and coronary heart disease in postmenopausal women. Atherosclerosis 216, 414419 (2011).

31. Kumawat, M. et al. Study of changes in antioxidant enzymes status in diabetic postmenopausal group of women suffering from cardiovascular complications. Clin. Lab. 58, 203207 (2012).

32. Mora, S. et al. Lipoprotein particle size and concentration by nuclear magnetic resonance and incident type 2 diabetes in women. Diabetes 59, 11531160 (2010).

33. Vekic, J. et al. Small, dense LDL cholesterol and apolipoprotein B: relationship with serum lipids and LDL size. Atherosclerosis 207, 496501 (2009).

34. Berger, J. S. et al. Lipid and lipoprotein biomarkers and the risk of ischemic stroke in postmenopausal women. Stroke 43, 958966 (2012).

35. Mattsson, C. & Olsson, T. Estrogens and glucocorticoid hormones in adipose tissue metabolism. Curr. Med. Chem. 14, 29182924 (2007).

36. Santosa, S. & Jensen, M. D. Adipocyte fatty acid storage factors enhance subcutaneous fat storage in postmenopausal women. Diabetes 62, 775782 (2013).

37. Maynar, M. et al. Menopause-induced changes in lipid fractions and total fatty acids in plasma. Endocr. Res. 27, 357365 (2001).

38. Milewicz, A., Tworowska, U. & Demissie, M. Menopausal obesity: myth or fact? Climacteric 4, 273283 (2001).

39. Sorci-Thomas, M. G. & Thomas, M. J. Why targeting HDL should work as a therapeutic tool, but has not. J. Cardiovasc. Pharmacol. 62, 239246 (2013).

40. Felig, P., Marliss, E. & Cahill, Jr G. F. Plasma amino acid levels and insulin secretion in obesity. N. Engl. J. Med. 281, 811816 (1969).

41. Carobbio, S. et al. Insulin secretion proles are modied by overexpression of glutamate dehydrogenase in pancreatic islets. Diabetologia 47, 266276 (2004).

42. Greeneld, J. R. et al. Oral glutamine increases circulating glucagon-like peptide 1, glucagon, and insulin concentrations in lean, obese, and type 2 diabetic subjects. Am. J. Clin. Nutr. 89, 106113 (2009).

43. Samocha-Bonet, D. et al. Glutamine reduces postprandial glycemia and augments the glucagon-like peptide-1 response in type 2 diabetes patients.J. Nutr. 141, 12331238 (2011).44. Cheng, S. et al. Metabolite proling identies pathways associated with metabolic risk in humans. Circulation 125, 22222231 (2012).

45. Lomivorotov, V. V. et al. Glutamine is cardioprotective in patients with ischemic heart disease following cardiopulmonary bypass. Heart Surg. Forum 14, E384E388 (2011).

46. Sut, A. et al. Pharmacologically dosed oral glutamine reduces myocardial injury in patients undergoing cardiac surgery: a randomized pilot feasibility trial. JPEN J. Parenter. Enteral. Nutr. 36, 556561 (2012).

47. Qi, L. et al. Association between a genetic variant related to glutamic acid metabolism and coronary heart disease in individuals with type 2 diabetes. JAMA 310, 821828 (2013).

48. Xu, F. et al. Metabolic signature shift in type 2 diabetes mellitus revealed by mass spectrometry-based metabolomics. J. Clin. Endocrinol. Metab. 98, E1060E1065 (2013).

49. Wrtz, P. et al. High-throughput quantication of circulating metabolites improves prediction of subclinical atherosclerosis. Eur. Heart J. 33, 23072316 (2012).

50. Hu, F. B. et al. Fish and omega-3 fatty acid intake and risk of coronary heart disease in women. JAMA 287, 18151821 (2002).

51. Imamura, F. et al. Novel circulating fatty acid patterns and risk of cardiovascular disease: the Cardiovascular Health Study. Am. J. Clin. Nutr. 96, 12521261 (2012).

52. Hu, F. B., Cho, E., Rexrode, K. M., Albert, C. M. & Manson, J. E. Fish and long-chain omega-3 fatty acid intake and risk of coronary heart disease and total mortality in diabetic women. Circulation 107, 18521857 (2003).

53. Lorente-Cebrin, S. et al. Role of omega-3 fatty acids in obesity, metabolic syndrome, and cardiovascular diseases: a review of the evidence. J. Physiol. Biochem. 69, 633651 (2013).

54. Lopez-Huertas, E. The effect of EPA and DHA on metabolic syndrome patients: a systematic review of randomised controlled trials. Br. J. Nutr. 107, S185S194 (2012).

55. Brouwer, I. A., Wanders, A. J. & Katan, M. B. Trans fatty acids and cardiovascular health: research completed? Eur. J. Clin. Nutr. 67, 541547 (2013).

56. Childs, C. E., Romeu-Nadal, M., Burdge, G. C. & Calder, P. C. Gender differences in the n-3 fatty acid content of tissues. Proc. Nutr. Soc. 67, 1927 (2008).

57. Yamatani, H., Takahashi, K., Yoshida, T., Soga, T. & Kurachi, H. Differences in the fatty acid metabolism of visceral adipose tissue in postmenopausal women. Menopause 21, 170176 (2014).

58. Hutchins-Wiese, H. L. et al. The impact of supplemental n-3 long chain polyunsaturated fatty acids and dietary antioxidants on physical performance in postmenopausal women. J. Nutr. Health Aging 17, 7680 (2013).

59. Song, Y. et al. Intake of small-to-medium-chain saturated fatty acids is associated with peripheral leukocyte telomere length in postmenopausal women. J. Nutr. 143, 907914 (2013).

60. Sammel, M. D., Freeman, E. W., Liu, Z., Lin, H. & Guo, W. Factors that inuence entry into stages of the menopausal transition. Menopause 16, 12181227 (2009).

61. Sherman, B. M., West, J. H. & Korenman, S. G. The menopausal transition: analysis of LH, FSH, estradiol, and progesterone concentrations during menstrual cycles of older women. J. Clin. Endocrinol. Metab. 42, 629636 (1976).

62. Tukiainen, T. et al. Detailed metabolic and genetic characterization reveals new associations for 30 known lipid loci. Hum. Mol. Genet. 21, 14441455 (2012).

63. Yu, Z. et al. Differences between human plasma and serum metabolite proles. PLoS ONE 6, e21230 (2011).

64. Ala-Korpela, M., Kangas, A. J. & Soininen, P. Quantitative high-throughput metabolomics: a new era in epidemiology and genetics. Genome Med. 4, 36 (2012).

65. Vartiainen, E. et al. Thirty-ve-year trends in cardiovascular risk factors in Finland. Int. J. Epidemiol. 39, 504518 (2010).

66. Viikari, J. et al. Serum lipids and lipoproteins in children, adolescents and young adults in 1980-1986. Ann. Med. 23, 5359 (1991).

10 NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2014 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5708 ARTICLE

67. Kaprio, J. The Finnish Twin Cohort Study: an update. Twin Res. Hum. Genet. 16, 157162 (2013).

68. Eriksson, J. G. et al. Catch-up growth in childhood and death from coronary heart disease: longitudinal study. BMJ 318, 427431 (1999).

69. Viechtbauer, W. Conducting meta-analyses in R with the metafor package.J. Stat. Softw. 36, 148 (2010).70. Patterson, H. D. & Thompson, R. Recovery of inter-block information when block sizes are unequal. Biometrika 58, 545554 (1971).

Acknowledgements

This work was supported by the Finnish Medical Foundation and Paulo Foundation. M.P. is partly nancially supported for this work by the Finnish Academy SALVE programme Pubgensense 129322 and by grants from Finnish Foundation for Cardiovascular Research. Metabolomic measurements of FINRISK 1997 were funded by the grant no. 139635 from the Academy of Finland to V.S. V.S. was also supported by the Finnish Foundation for Cardiovascular Research. The study has been supported by the Academy of Finland (P.S.), TEKESthe Finnish Funding Agency for Technology and Innovation (M.A.-K.), the Sigrid Juselius Foundation (M.A.-K.), the Strategic Research Funding from the University of Oulu (M.A.-K.) and by the European Union Seventh Framework Programmeproject BioSHaRE (FP7/2007-2013) grant no. 261433. A.J.is funded by the Helsinki Biomedical Graduate Program and the Helsinki Doctoral Program in Clinical Research. The work has also been funded by the European Foundation for the Study of DiabetesEFSD (M.A.-K., A.M.). The Young Finns Study has been nancially supported by the Academy of Finland: grants 134309 (Eye), 126925, 121584, 124282, 265240, 263278, 264146, 129378 (Salve), 117797 (Gendi) and 41071 (Skidi), the Social Insurance Institution of Finland, Kuopio, Tampere and Turku University Hospital Medical Funds, Juho Vainio Foundation, Paavo Nurmi Foundation,

Finnish Foundation of Cardiovascular Research and Finnish Cultural Foundation, Tampere Tuberculosis Foundation and Emil Aaltonen Foundation.

Author contributions

K.A. and A.J. were responsible for the study design, statistical analyses, data interpretation and writing the manuscript. K.F. was responsible for the analyses of Estonian data and assisted in designing the study and writing the manuscript. J.Ke., H.M. and M.N. participated to data analyses. P.S., A.J.K. and M.A.-K. designed and performed the NMR analyses. J.Ka., T.L., O.R., J.G.E., V.S. and A.M. contributed to study cohort collection and metabolite analyses. M.A.-K., A.S.H., M.J. and A.T. participated in data interpretation and study design. M.P. supervised the study and participated in study design and data interpretation. All authors contributed to writing and reviewing the manuscript.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: P.S., A.J.K. and M.A.-K. are shareholders of Brainshake Ltd, a startup company offering NMR-based metabolite proling. K.A. is currently employed by GSK. The remaining authors declare no competing interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Auro, K. et al. A metabolic view on menopause and ageing. Nat. Commun. 5:4708 doi: 10.1038/ncomms5708 (2014).

NATURE COMMUNICATIONS | 5:4708 | DOI: 10.1038/ncomms5708 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 11

& 2014 Macmillan Publishers Limited. All rights reserved.